Developing and Validating a Language Model to Extract Adult Emergency Department Records for Clinical Research from 2021 to 2023

What this Trial Is About

Researchers are developing and validating a language model called eCREAMLM designed to interpret emergency department electronic health records EHRs from adult patients treated between 2021 and 2023. This observational, retrospective study aims to see if the language model can accurately extract relevant information from free-text medical notes for research and clinical analysis across six languages. The study involves multiple centers across several European countries and addresses challenges in emergency medicine research due to high patient volume and limited staff for data collection. The eCREAMLM model will be created by training and fine-tuning existing open-source language models on billions of medical texts and millions of free-text notes from hospital records. Expert physicians will annotate thousands of clinical notes to guide the fine-tuning process. Once developed, the models accuracy will be validated by comparing its data extraction against physician annotations in 1,000 clinical notes. Data will be anonymized and pooled by language to protect patient privacy. Participants contribute by having their anonymized emergency department medical notes included in the study database. The researchers will analyze how well the language model fills out a virtual case report form compared to experts, using statistical measures of agreement. This process requires no active participation from patients. The study is expected to run until September 2027, with outcomes measured at one month after validation starts.

Development of a Natural Language Processing Tool to Enable Clinical Research in Emergency Medicine